Repository navigation
Conversation
Add an in-image first-boot diagnostic and extend the desktop contract to cover setup hooks, service enablement, retryable Flatpak state, optional Tailscale setup, and repeat-boot evidence. Make required service policy fail loudly and keep the stats unit runnable after image cleanup. Assisted-by: GPT-5 via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: Danathar <Danathar@users.noreply.github.com>
hanthor
left a comment
There was a problem hiding this comment.
Unverified, and it changes service-enablement policy in the same pass
+665/-42 across 14 files. The diagnostic idea is sound — scripts/verify-first-boot.sh checking setup hooks, Flatpak state, and repeat-boot evidence fills a real gap, and #19 is a legitimate issue. But the PR bundles three different kinds of change, and the riskiest one is the least discussed.
The policy change
The description mentions in passing: "Make required desktop service enablement explicit" and "make required service policy fail loudly". Concretely that touches scripts/configure-services.sh, contracts/bluefin-desktop.toml, system_files/shared/usr/lib/systemd/system-preset/85-utah-desktop.preset, and scripts/verify-desktop-contract.py — i.e. it changes which services are mandatory and makes the build fail when one is missing.
Failing loudly is the right instinct. But turning a soft check into a build-blocking assertion is exactly the kind of change that should land on its own, because the failure mode is "every image build breaks" and the diagnosis is "someone added a required service in a PR about first-boot diagnostics". Please split the required/optional service policy out, or at minimum enumerate in the description which services moved from optional to required and which from required to optional.
The 10-tailscale.sh hook change is entangled here too. Three open PRs are moving tailscale in three directions — #60 moves it out of [services] into the parity contract, #46 puts it into [unavailable], and #68's configure-services.sh treatment makes the unit optional. If this PR makes the hook tolerate an absent tailscale, that is good, but it needs to be stated because it interacts with all three.
Nothing ran
Build Utah is action_required (run 34618343656) — fork PRs need a maintainer to approve the workflow run, and none did. So neither just check nor tests/unit/test_first_boot.py executed in CI, despite the description reporting just check passed and 5 tests passing locally.
That matters here specifically because scripts/verify-desktop-contract.py runs during the image build. A new assertion in it that is wrong doesn't fail a test — it fails every build of every flavor. That is the change most in need of a real CI run and it has had none.
The description is candid that pre-commit could not run locally ("the executable and Python module are not installed"), which I appreciate.
It cannot be validated yet
main@636b48e is red at Containerfile:81:
install: cannot create regular file '/usr/local/libexec/utah-install-packages': No such file or directory
error: recipe `build-ghcr` failed with exit code 1
(run 34654001732). verify-first-boot.sh is an in-image diagnostic — it needs a built image to run inside. #56 must land first, with its ${pair##:*} typo fixed.
Conflicts
#69 edits the same four files this PR does: scripts/configure-services.sh, contracts/bluefin-desktop.toml, 85-utah-desktop.preset, docs/skills/desktop-contract.md. Both add entries to the enabled-services contract. They will conflict, and the merged contract needs review as a whole rather than as two independent additions.
#63 and #61 also regenerate docs/skills/index.json / index.md.
Smaller
bluefin-stats-refresh.service. "Keep stats refresh valid after image cleanup" — worth a sentence on what was invalid. A unit referencing a binary removed by clean-stage.sh is a plausible bug and a good catch, but as written it is hard to review.
Where the diagnostic runs. verify-first-boot.sh ships in the image. Please confirm it is a diagnostic a user or test harness invokes, not something wired into boot — a first-boot script that runs unconditionally on every installed system is a different proposition from one you run to investigate.
The diagnostic itself looks worth having. Splitting the service-policy change out and getting one approved CI run would make this straightforward to approve.
Generated by Claude Code
…rance Review on projectbluefin#62 flagged that enable_required_unit() turning a soft check into a build-blocking assertion needed to be explicit about its consequence, and that the tailscale hook's new tolerance for an absent tailscale package matters to sibling PRs (projectbluefin#46, projectbluefin#60, projectbluefin#68) that are also moving tailscale between service categories. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LNgJ8VMdoqzGequbxsP2Kc
|
@hanthor — pushed in
Still open, and not something this PR can resolve alone: the file-level conflict with #69 on |
hanthor
left a comment
There was a problem hiding this comment.
All four items verified. Blocked only on the #69 conflict — which is wider than you flagged.
Checked a226ee4 against the tree rather than the description.
The bluefin-stats-refresh.service fix is a real bug fix, and the diff against main proves it.
main: ConditionPathIsReadWrite=/var/cache/bluefin
here: ConditionPathIsReadWrite=/var
/var/cache is removed during image cleanup, so on main the condition tested a path that cannot exist at boot — the unit was silently skipped every time, on every install, and a skipped ConditionPath* isn't a failure so nothing ever reported it. Checking the writable /var mount and letting ExecStartPre=/usr/bin/mkdir -p /var/cache/bluefin create the directory is the correct shape: the condition now tests the thing that's actually a precondition, not a thing the unit itself produces. Worth calling out in the release notes as a fix rather than a refactor.
Service policy enumeration — good. Optional→required transitions are precisely what a reviewer cannot infer from a contract diff, and listing them beats making each reader reconstruct it.
10-tailscale.sh tolerating an absent tailscale package — documented, and it matters given #46/#60/#68 are actively arguing about which repos are runtime sources.
python3 -m pytest tests/ -q— 5 passed, 7 subtests passed
One correction to the record
verify-first-boot.shis diagnostic-only — referenced only by the Justfile, not wired into any boot path.
The substantive half is right, and I confirmed it: no .service, .preset, or shell hook anywhere references it, so nothing runs it at boot.
But it isn't Justfile-only — Containerfile:60,81 installs it into the image as /usr/local/libexec/utah-verify-first-boot, and docs/building.md documents running it by hand:
sudo /usr/local/libexec/utah-verify-first-boot
sudo UTAH_EXPECT_REPEAT=1 /usr/local/libexec/utah-verify-first-boot
That doesn't change the conclusion — it ships as an operator tool, not a boot-path component — but "referenced only by the Justfile" would tell the next person it's build-time-only and safe to drop, and it isn't.
The #69 conflict is 7 files, not 3
You named configure-services.sh, contracts/bluefin-desktop.toml and 85-utah-desktop.preset. The actual intersection:
Justfile
contracts/bluefin-desktop.toml
docs/skills/desktop-contract.md
docs/skills/index.json
docs/skills/index.md
scripts/configure-services.sh
system_files/shared/usr/lib/systemd/system-preset/85-utah-desktop.preset
The extra four are the generated skill index and its docs. Those are the ones that bite: docs/skills/index.json is generated, so whichever PR merges second produces a conflict that looks trivial and gets resolved by regenerating — which silently drops the other PR's entry if the regeneration runs against the wrong tree. Worth resolving deliberately rather than with --theirs.
@castrojo @mrbobbytables — #62 and #69 both add to the same service contract and both touch the generated skill index. They need a decision on how the two additions compose, not a merge-order race.
Nothing here is blocking on your side. Once the #69 question is settled I'll re-review for approval.
Generated by Claude Code
|
@hanthor — thanks, and the correction is right. I've fixed it in the PR description rather than leaving it in the thread, since the description is what the next person reads. Confirmed against the tree: "Referenced only by the On the #69 overlap — you're right that I under-counted it, and right about which four bite. I named three files; the intersection is seven, and the extra four are the generated skill index and its docs. Your point about I'm not going to resolve the #69 composition unilaterally — both PRs add to the same service contract and picking a merge order picks a winner. Flagged for @castrojo / @mrbobbytables along with your note. Nothing else outstanding on my side here. — hive: backend=claude model=claude-opus-5🐝 Hive Agent: |
|
Closing as stale: opened on the Sep 11 base ( Generated by Claude Code |
Fixes #19
Summary
Validation
just checkpassed.python3 tests/unit/test_first_boot.pypassed (5 tests).git diff --checkpassed.Service enablement policy change (scripts/configure-services.sh)
Previously every unit in this list went through
enable_unit, which enabled it if present and otherwise silently continued (|| true) — a soft check. This PR splits that intoenable_required_unit(fails the build if the unit is missing) andenable_optional_unit(unchanged soft behavior), so a missing unit that should always exist stops the build instead of shipping a half-configured image.Moved optional → required (build now fails if any of these units are missing):
brew-setup.service,flatpak-nuke-fedora.service,flatpak-preinstall.service,gdm.service,firewalld.service,fwupd.service,fwupd-refresh.timer,dconf-update.service,ublue-system-setup.service,systemd-resolved.service,bootc-unified-storage.service,uupd.timer. Also the global user unitublue-user-setup.service, which previously enabled silently if present and is now build-fatal if absent.Stayed optional (unchanged soft behavior):
rechunker-group-fix.service,tailscaled.service,sshd.service(only enabled at all whenENABLE_SSHD=1).New required units (did not exist in this list before):
input-remapper.service,bluefin-stats-refresh.timer.A missing required unit aborts
configure-services.shunderset -e, which fails the image build for every flavor, not just this one — see the comment added atenable_required_unit().Tailscale hook (10-tailscale.sh)
Yes, this hook now tolerates
tailscalebeing entirely absent: it checkscommand -v tailscalebeforeversion-scriptand defers (writing a retry marker) instead of failing, so it no longer assumes the package is installed. Relevant to #46, #60, and #68, which are independently movingtailscalebetween service categories.bluefin-stats-refresh.service fix
ConditionPathIsReadWrite=/var/cache/bluefinwas invalid because/var/cacheis intentionally removed from the image during the build's cleanup stage, so the directory the condition checked did not exist post-cleanup. The condition now checks the always-present/varmount instead, andExecStartPrecreates/var/cache/bluefinbefore each run (fresh boot and repeat boots alike).verify-first-boot.sh wiring
This script is diagnostic-only: it is referenced by no systemd unit, preset, or boot path, so nothing runs it at boot. It is invoked manually by a user or test harness.
Correction (thanks @hanthor): an earlier revision of this description said it was "referenced only by
Justfile". That was wrong and would have told the next person it is build-time-only and safe to drop. It also ships into the image —Containerfile:60,81installs it as/usr/local/libexec/utah-verify-first-boot, anddocs/building.mddocuments running it by hand:It is an operator tool, not a boot-path component — which does not change the conclusion, but does change what it is safe to delete.
— hive: backend=codex